An evaluation metric in Universal Grammar provides a means of selecting between possible grammars for a particular language. The evaluation metric as conceived in Chomsky & Halle (1968; henceforth SPE) prefers the grammar in which only the idiosyncratic properties are lexically listed and predictable properties are derived. The essence of underspecification theory is to supply such predictable distinctive features or feature specifications by rule. Viewed in this way, the general idea of underspecification has always been a part of any theory of phonology that includes such an evaluation metric.

One of the common assumptions of standard metrical phonology (Hayes 1981) is that rules of stress placement do not refer to the weight or nature of syllable onsets. This is most clearly stated in Halle & Vergnaud (1980:93): ‘in all languages known to us, stress assignment rules are sensitive to the structure of the syllable rime, but disregard completely the character of the onset’. This assumption has been attacked by both Davis (1982, 1985a, b) and Everett & Everett (1984). These researchers have pointed to a number of languages which seem to have stress-placement rules that are sensitive to the nature of the syllable onset. For example, Davis (1985a, b) notes that Western Aranda (an Arandic language of Australia) has a stress rule that places main stress on the initial syllable if it begins with a consonant; otherwise stress falls on the second syllable. However, if the word is bisyllabic, stress must fall on the initial syllable, since final syllables are always extrametrical.

The theory proposed here, the Markedness Theory of Syllable Structure (MTSS), provides an account of syllabic phonotactics, wherein not only are features defining phonological content underspecified, but also those which determine the number and order of segments. The descriptive basis of MTSS in this paper consists of the minimally redundant underlying representations (URs) of stressed syllables in English. These forms are parsimoniously accounted for by a theory in which content features are associated with prosodic nodes in UR, and which contains an algorithm which maps UR prosodic nodes, specified for feature content, into strings of timing units (x's on the skeletal tier), with fully specified syllabic structures on the prosodic tier.

The West African language Kasem, which is spoken by the Kasenas on both sides of the northern border between Ghana and Burkina Faso, figured prominently in the abstractness controversy in the late seventies. The analyses proposed by Halle (1978) and Phelps (1975, 1979), in particular, show most clearly that the framework of linear phonology does not provide a suitable model for the description of the complex set of alternations that can be observed in the nominal system of this language. I will argue that the theory of underspecification (cf. Kiparsky 1982, 1985; Archangeli 1984; among others) together with the principles of CV phonology (cf. McCarthy 1979, 1981; Clements & Keyser 1983; Levin 1985) provide the necessary equipment to account for the Kasem facts in a relatively simple and insightful way. It will turn out that these subtheories enable us to collapse several seemingly independent processes into a rather small set of language-specific rules.

Over the past few years, research in metrical phonology has witnessed a shift in its main topic of investigation. Originally, attention was focused on the representation of prominence patterns of words (for example, Liberman & Prince 1977; Kiparsky 1979; Selkirk 1980; Hayes 1981), but more recently interest has arisen in several sorts of ‘rhythmic’ stress phenomena in larger domains (Prince 1983; Hayes 1984; Selkirk 1984; Hammond 1984; Giegerich 1985). One way of explaining this shift is by noting that the issue of the treatment of prominence patterns proper seems to have reached a stage where both grid-only (i.e. tree-less) theory and variants of tree-full theories, whether or not they employ grids as well, are capable of explaining the prominence patterns of words (see, for instance, Prince 1983; van der Hulst 1984). In this situation, investigation of rhythmic stress phenomena may offer the possibility of evaluating these theories because, as currently perceived, this area typically deals with the issue of whether metrical tree structure is needed at all, or whether grid structure by itself is capable of explaining rhythmic adjustments: on the one hand, grid-only theory claims that any constituency relevant to rhythmic adjustment is adequately encoded in the grid; on the other, tree theory holds that more detailed, or perhaps different, constituent information is required, as expressed in the metrical tree.

Since the advent of distinctive feature theory, few issues have received as many interpretations as the phonological representation of vowel height. Vowel height features have been denned acoustically and articulatorily, have allowed three, four or five distinct heights, have been unary, binary and n-ary, and have been on a single tier, multiple tiers or in various head-dependency relationships. It is fair to say that there is no consensus on how vowel height should be represented. While many generative phonologists have been quite content working for nearly three decades with a pair of binary vowel height features, the literature of this period includes a steady flow of criticisms of this approach as well as suggestions for improvement or radical change. This literature generally addresses itself to two problems inherent in the SPE features [high] and [low] and the three vowel heights they define: First, how does one account for systems with four (five?) vowel heights? Second, how can rules that raise (lower) vowel heights by one step each be accounted for? In the first question we ask what the theory has to say about distinctive oppositions such as /i e ∊ æ/ in the second question we ask what the theory has to say about a rule such as one that would lower /i/ to [e] and /e/ to [∊].

The phenomenon studied in this paper is the correspondence between the syllabic position of segments copied in reduplication and the syllabic position of their base counterparts. I will document this correlation and propose a model of reduplication that explains it.

It is often assumed in work on the phonological underspecification of segments that while representations may at first be underspecified, they end up fully specified. Various kinds of rules are posited to ensure that in output forms all features have values for all segments. In this paper I will consider an alternative view: that underspecification may persist into phonetic representations. I begin by reviewing some of the relevant phonological phenomena and mechanisms, and some of the history of underspecification in phonetic studies. I then show how phonetic data may be taken to reflect the presence or absence of feature values in surface forms. Finally, the extent to which surface specification depends on segmental contrasts is examined. The aim of this paper is to present some phonetic phenomena that are potentially relevant to any theory of underspecification.

Shattuck-Hufnagel (1986) (henceforth S-H), in discussing the significance of ‘slips of the tongue’ for a model of processing, is particularly concerned with attempting to establish to what extent some of the constructs posited by phonologists in their characterisations of phonological structure can be shown to have a role in the planning of speech production. On the basis of the ‘MIT corpus’ of errors (collected by Merrill Garrett and herself), particularly those involving vowels, she argues for a role in planning for aspects of syllable structure, placement of lexical stress, and distinctions and dimensions such as those commonly captured by feature notations. My concern here is with some further aspects of syllable structure: I shall suggest that her data provide support for even more detailed structural properties than are proposed by S-H herself; indeed, that one uncertainty in assessing the significance of the data can be resolved by an independently motivated analysis of subsyllabic structure. I shall not repeat here the details of the classification of errors on which S-H's discussion is based (1986: §1); what is significant in the present context is that errors of various types provide a motivation for particular subsyllabic units.

The mettle of approaches to synchronic linguistics can often be profitably tested by confrontation with language patterns from paralinguistic areas. In the following pages underspecification theory will be shown to play a central role in illuminating the nature of Turkish rhyme in the classical tradition, where the conditions for euphonic matching of vowels otherwise elude neat linguistic categorisation. As a by-product, the analysis also provides some clarification of a detail of suffix allomorphy.

The major issue with regard to underspecification theory is whether all types of redundant information should be excluded from underlying representations, or if not, then what the principles are that determine the inclusion or exclusion of particular types of information. The vowel system of Tiv, a Niger-Congo language of Nigeria, is particularly interesting in this regard since the representation of individual vowels in verbal roots requires a type of radical underspecification: no features are actually assigned to vowels underlyingly; surface forms result from the interaction of morpheme-level specifications with rules of spreading and redundancy.

Phonologists have known for some time that the so-called ‘standard’ theory of generative phonology is not adequate for the analysis of vowel harmony. Ringen (1975, 1977, 1980) suggests that some of the problems can be solved by abandoning the assumption that phonological representations are fully specified. Clements (1977b, 1980) suggests that vowel harmony should be analysed autosegmentally. Underspecification theory, developed in the recent work of Kiparsky, Archangeli and Pulleyblank, incorporates both of these proposals. This paper considers how Hungarian can be analysed within this theory. It is shown that by adopting Goldsmith's (1985) proposal that vowel harmony in Hungarian involves the spreading of the feature [−back], the transparent (neutral) vowels in Hungarian are derived because the redundancy rule assigning [−back] to these vowels, although available, does not apply early in the derivation because its structural description is not met.

The recent phonological literature has witnessed the emergence of a significant body of research under the rubric of underspecification theory. This model of inquiry was first proposed by Kiparsky (1982), developed more deeply in Archangeli (1984), and is articulated most exhaustively in Archangeli & Pulleyblank (forthcoming a). One of the most basic assumptions advanced in these works is that for each contrastive feature one value is specified underlyingly and the other is inserted by default. Feature changing harmony, which requires both values to be underlying, is a particularly interesting challenge to this claim and appears to undermine its restrictiveness. Indeed, it figures prominently in some recent proposals, such as Steriade (1987b), to justify relaxing the theory and admit both feature values at the underlying level.